SHARE:

AI in the help desk is no longer talk about the future. It already shows up in government purchasing plans. The US Department of Veterans Affairs, the VA, is running market research to bring its enterprise service desk under a single managed services contract. The contract would last five years, with an artificial intelligence automation plan built in from the first draft. The news comes from OrangeSlices AI, a site that tracks the US public procurement market, in a report published on November 19, 2026.

A quick heads-up first: there is no signed contract yet. What the VA opened was a request for information, the well-known RFI. This is the stage where an agency talks with vendors before deciding how it will buy.

Even so, the size of the operation is not the most interesting part. The approach is. Instead of switching on AI all at once and hoping for the best, the agency designed a gradual reduction of human support capacity over the years. And no new automation feature talks to a real user without approval first.

That is the lesson for any company, big or small: service desk automation tends to work better when it moves forward a little at a time. Below, you will see the VA plan and a practical, step-by-step path to bring AI to your IT support without messing up the routine of the people who depend on it.

What the VA is planning for the service desk

According to OrangeSlices AI, the VA wants a single managed services contract for its enterprise service desk. Right now, Tier 0 and Tier 1 support and the contact center platform services are split across two separate task orders. The idea is to put everything under one roof, with one party in charge of the operation.

If those terms sound strange, here is the plain version. Tier 0 is the self-service level, when users fix the problem themselves with a portal, a help article or a virtual assistant. Tier 1 is the first line of human support, the person who picks up the ticket and solves the most common issues. Putting both under the same management makes sense precisely because AI will push contacts from one level to the other.

The scale is impressive. According to the report, the service desk supports about 600,000 employees and contractors and gets around 126,000 contacts a month. Of those, roughly 95,000 come in by phone. The winning vendor would have to provide support 24 hours a day, every day of the year, run the cloud contact center platform and handle the knowledge base, process integration and connections with the IT service management tool, plus quality control and performance reporting.

The report also says the environment has to handle peaks of up to 2,500 calls in 15 minutes and up to 550 licenses in use at the same time. That is no small thing.

There is one detail that is easy to miss, but it is worth a lot. The VA currently uses NICE CXone for its contact center and ServiceNow for IT service management, according to OrangeSlices AI. Still, the requirement was written to be technology neutral, and the agency may switch platforms while the contract is running. In other words, the focus is on results, not on the brand of the tool.

According to the outlet, the RFI is meant to assess market capability, transition feasibility, pricing, contract structure, possible small business set-asides and barriers to competition. It is the kind of homework many companies skip when they decide to buy technology.

The automation ramp: from 92% to 60% human support

The heart of the plan is what the report calls a glide path, a bit like a landing approach. Under the proposal described by OrangeSlices AI, contracted human support capacity starts at 92% of a certified baseline in year one and drops to 60% by year five. Suitable contacts gradually move from Tier 1 agents to self-service, automation and AI-powered channels.

Receive the best innovation content in your email.

All the news, tips, trends, and resources you're looking for, delivered to your inbox.

By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.

Notice the word suitable. The plan does not say everything goes to the bot. It says whatever makes sense to automate gets automated, at a predictable pace.

The technology menu is broad. According to the outlet, it includes conversational AI, generative AI, agentic AI, robotic process automation (RPA) and agent assist tools. It is not a bet on a single type of solution but a mix, with each piece doing the job it does best. Conversational AI chats, generative AI writes, agentic AI carries out tasks, RPA repeats processes across systems and agent assist supports the person on the other end of the line.

Now for the part that matters most if you are thinking about doing something similar: the VA requires prior approval before any new automation or AI feature handles a real contact. Nothing goes live on a whim. That emergency brake is what makes an ambitious plan safer to carry out.

Why a phased AI rollout makes sense

The first reason is simple: less risk for the people who need help. In internal support, the user is your employee. If they get stuck with a bot that does not understand the request, they lose time, get annoyed and hold up other people as well. Moving in phases limits the damage when something does not go as planned.

Picture an illustrative example. A company decides to replace all of its support with a chatbot on a Friday. On Monday, the finance system goes down, dozens of people open tickets at the same time, and the bot only knows how to answer password questions. Nobody can reach a technician. When the change happens in stages, that kind of scene becomes much less likely.

The second reason is time to learn. Each stage of a phased AI rollout produces information about what works, what confuses people and what needs tweaking. That window also helps train the team, who start working alongside the technology instead of being caught off guard by it. The conversation about the impact on jobs is real, and it is worth reading about how automation affects jobs to handle this transition with care.

The third reason is having something to compare against. The VA starts from a certified baseline, and that makes all the difference. Without knowing how support performs today, there is no way to say whether AI improved anything. With a clear starting point, every step forward can be measured and justified to leadership.

How to bring artificial intelligence to IT support at your company

You do not need 600,000 users to use the VA logic. The same thinking works for a team of 50 people or a company with several branches. What changes is the size of each step.

Artificial intelligence in IT support tends to pay off more when it comes in layers. First the groundwork, then the basics, then support for agents and, finally, full automation.

Here is how that can look in practice.

Step 1: get your house in order

Before any bot, review your knowledge base. Outdated articles, procedures that only live in one technician’s head and conflicting answers will feed the AI bad information. And AI with bad information tends to give wrong answers with a lot of confidence.

Next, clean up your ticket categories. If a big chunk of tickets ends up in a catch-all other bucket, you do not know what repeats the most, and repetitive work is exactly what automation handles best.

Finally, record your current metrics: volume, resolution time, satisfaction. That is your baseline, just like the VA did.

Step 2: self-service for the basics

Start with what is boring, frequent and low risk. Password resets, system access requests, VPN setup, printer questions. These requests eat up the team all day and almost never need deep analysis.

A well-trained conversational assistant can solve many of these cases on its own, at any hour. Picture an employee who forgets their password at seven on a Monday morning and can unlock the account without waiting for anyone to get to the office.

Always make it clear what the assistant can and cannot do. Setting the right expectations prevents frustration.

Step 3: AI as the agent copilot

Here, AI does not talk to the user. It helps the person who does. It suggests replies based on the knowledge base, summarizes long tickets and handles initial triage, flagging priority and the team responsible.

This is a great stage for building trust. The agent reviews everything before sending, so AI mistakes get caught along the way and become lessons. The team also starts to see the technology as an ally, not a threat.

This is a good moment to work on observability, which means being able to see what the AI is doing and why. We have a piece here on the blog about observability and automation in AI-powered support that goes deeper into the topic.

Step 4: end-to-end automation

Once the earlier steps are running smoothly, you can automate entire workflows. An access request can be received, approved by the manager, carried out in the system and closed without a technician ever touching it. RPA and AI agents come in right here.

The golden rule is the VA rule: nothing goes live without approval. Test with a small group, check the results, get sign-off from whoever owns the area and only then roll it out to everyone.

Tools we use daily

If you want ideas for designing these processes, take a look at the piece on AI and automation in workflows.

How to keep an AI-powered internal help desk from becoming a headache

The first precaution is the emergency exit. Every AI-powered internal help desk needs an easy way to reach a person. If employees have to guess the magic word to get transferred, the experience falls apart fast. In the VA plan, human capacity shrinks, but it is still there.

The second is preparing for peaks. The VA itself treats demand spikes as a requirement, according to OrangeSlices AI. At your company, think about system outages, tool changes or month-end close. Have a plan B for when automation fails or volume explodes, with people available to hold the line.

The third is avoiding vendor lock-in. The technology-neutral model of the VA teaches this well. Document your processes, keep the knowledge base exportable and favor solutions that work well with others.

And do not forget data protection laws, such as the LGPD, the Brazilian General Data Protection Law. IT tickets carry names, job titles, access rights and sometimes very sensitive information. Define what the AI can read, where that data is stored and who has access to it.

How to tell if service desk automation is working

Without numbers, the conversation turns into guesswork. That is why you should track a small set of indicators from day one and always compare them with the baseline from Step 1.

  • Self-service resolution rate: how many requests get solved with no human involvement.
  • Average resolution time: whether AI is speeding up support or just moving the bottleneck somewhere else.
  • Internal satisfaction: a short survey at the end of each ticket already helps a lot.
  • Volume still reaching agents: the equivalent of the VA ramp, showing how much work has actually shifted.

Watch out for mixed signals. If self-service resolution goes up but satisfaction goes down, the bot may be closing tickets it did not really solve. Reading the numbers side by side keeps you from celebrating too early.

One more tip: set goals for each phase, just like the VA did with its ramp. Instead of promising to transform all of support, agree with the team on what changes at each stage and when you will check your course. Adjusting becomes much easier and less dramatic.

The VA case shows you can think big without skipping steps: clear goals for each phase, approval before every leap and people at the center of the process. A good place to start is to look at your help desk today, set your baseline and pick one repetitive request to automate first. The next steps get much clearer after that, and the Método Viral content on automation and AI in business can help with your planning.

Picture of Rafael

Rafael

Operations

I transform internal processes into delivery machines — ensuring that every Viral Method client receives premium service and real results.

Fill out the form and our team will contact you within 24 hours.

Related publications

AI SDR Agent on WhatsApp: How SMBs Can Cut Costs and Scale Sales

Respond 21x faster your leads and scale your sales operation with a fraction of the cost of expanding your sales

Productivity with Agentic Artificial Intelligence in execution and workflows.

Agentic AI: how to operationalize AI agents to improve workflows, metrics, and governance, turning pilots into real productivity gains.

How AI Email Agents Are Boosting Productivity for Lean SME Teams

Learn How AI Email Agents Are Boosting Productivity for Lean SME Teams and get ahead of your competition

Receive the best innovation content in your email.

All the news, tips, trends, and resources you're looking for, delivered to your inbox.

By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.

Rafael

Online

Atendimento

Website Pricing Calculator

Find out how much the ideal website for your business costs

Website Pages

How many pages do you need?

Drag to select from 1 to 20 pages

In just 2 minutes, automatically find out how much a custom website for your business costs

More than 500 companies have already calculated their quote

Fale com um consultor

Preencha o formulário e nossa equipe entrará em contato.